Robust retrieval augmented generation
Abstract
Retrieval Augmented Generation (RAG) enhances Large Language Models with external knowledge but is vulnerable to noisy and contradictory retrieved content. This study develops a modular framework for evaluating RAG robustness under controlled perturbations. Four QA benchmarks are normalised to a common schema and evaluated across two noise types, four intensity levels and six pipeline variants targeting different stages, producing 192 configurations and 96,000 evaluations. Results demonstrate that robustness gains are stage and noise type dependent: pre-retrieval interventions primarily address retrieval noise while context stage filtering most effectively mitigates contradiction noise. A consistent accuracy–faithfulness trade-off is observed, and combining all interventions produces non-additive interactions that can degrade calibration. Preliminary LLM-based validation confirms these trends transfer to generative pipelines. The findings show that no single intervention is universally optimal, motivating objective aware, stage specific robustness design for RAG systems.
Read the paper
Related studies
- Knowledge Management Perspective of Generative Artificial Intelligence
- Navigating Generative Artificial Intelligence Promises and Perils for Knowledge and Creative Work
- Characterizing the evolving research on enterprise content management
- Large language models and multimodal foundation models for precision oncology
- From Generative AI to Extended Reality: Multidisciplinary Perspectives on the Challenges, Opportunities, and Future of Educational Computing
- Generative Artificial Intelligence: A New Engine for Advancing Environmental Science and Engineering
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.